Deep learning-based crack detection method for civil engineering and its multidimensional damage assessment

Civil engineering cracks are located in complex and diverse environments, often interfered by a variety of external background factors, and crack detection faces various challenges. This paper proposes an improved YOLO v8s-WOMA network, which mainly introduces the ODConv module, C2f-MA module and WIoU loss function to solve the problem of many interferences and low accuracy of civil engineering crack identification in complex background. At the same time, combined with the determination criteria of civil engineering crack damage, the BP neural network is trained to evaluate the damage degree of cracks. Experimental validation is carried out on the CBP dataset dataset to compare the algorithm of this paper and with the existing target detection algorithm. The experimental results show that the proposed YOLO v8s-WOMA network outperforms other algorithms in several target detection performance evaluation indexes, and this paper’s algorithm has the highest mAP value, F1 value, and accuracy value, which are 90.5%, 90.3%, and 89.6%, respectively. The width error of the actual bridge crack detection does not exceed 0.1mm, the length error does not exceed 20mm, and it can accurately determine the degree of civil engineering damage. It shows that the model in this paper can effectively deal with the interference of complex background, accurately detect bridge cracks, and can basically meet the practical application of civil engineering crack detection.